Risk of Severe Sepsis After Acute Kidney Injury
Risk of Severe Sepsis After Acute Kidney Injury
The National Health Insurance (NHI) is a nationwide, compulsory, comprehensive health system in Taiwan. Patients were drawn from the NHI Research Database (NHIRD), which was released for research purposes by the National Health Research Institutes, Taipei, Taiwan. The NHIRD, one of the largest databases in the world, covers nearly all (99%) inpatient and outpatient claims for its population of more than 22 million people, and it has been used extensively in previous studies. The NHIRD provides encrypted patient identification numbers; age; sex; dates of hospital admission and discharge; medical institutions providing services; International Classification of Diseases, Ninth Edition, Clinical Modification (ICD-9-CM), codes of diagnoses and procedures; and outcome at hospital discharge (recovered, died or transferred out).
The NHIRD consists of all the original claim data and registration files of one million individuals from 1999 to 2008. These one million individuals were randomly sampled from the 2000 Registry for Beneficiaries of the Taiwan NHI program and screening comorbidities by using the medical records for one year preceding admission. We linked to diagnostic codes through the hospitalization claims data to identify all episodes of AKI and the corresponding order codes used for the study subjects. We defined the first AKI hospitalization requiring dialysis as the index hospitalization for each individual. Participants who received dialysis during the index hospitalization were included. We excluded individuals who had a previous diagnosis of AKI, had received a kidney transplant or any form of dialysis preceding the index hospitalization AKI (n = 6,420), were hospitalized for more than 180 days with AKI (n = 36), died during the index hospitalization AKI (n = 718) or withdrew from the NHI within three months after discharge (n = 1,382). Each AKI patient was then matched with four hospitalized patients without AKI according to age, sex and history of diabetes (Figure 1). Informed consent was originally obtained by the NHRI, and, since patients were anonymous in the present study, informed consent was not required. Also, since the identification numbers of all individuals in the NHRID were encrypted to protect the privacy of the individuals, this study was exempt from a full ethical review by the institutional ethics review board of National Taiwan University Hospital (reference no. 201212021RINC).
Patients were tracked for outcomes beginning 90 days after discharge and ending at 31 December 2009. The main study outcome was severe sepsis. The method of identification of patients with severe sepsis was similar to that used by Angus et al., who selected all acute care hospitalizations with ICD-9-CM codes for both a bacterial or fungal infection process and a diagnosis of acute organ dysfunction. We used codes for acute organ dysfunction as modified by Shen et al. (listed in Table 1). Sepsis was defined according to the American College of Chest Physicians/Society of Critical Care Medicine (ACCP/SCCM) as systemic inflammatory syndrome in response to infection, which, when associated with acute organ dysfunction, is said to be severe. For patients with more than one hospital admission for severe sepsis during the study period, the first episode of sepsis was included.
For individuals with and those without AKI and dialysis, continuous variables were compared using an unpaired t-test and are expressed as mean values with SD. Categorical variables were compared using a χ test and expressed as a percentage. The person-years were calculated from enrollment to outcome, death or the end of the study period. Incidence rates of severe sepsis were calculated for participants with and those without AKI and dialysis, respectively.
Each acute dialysis patient was matched with four non-AKI individuals according to age, sex and history of diabetes. Crude hazard ratios (HRs) with 95% confidence intervals (CIs) were derived from Cox proportional hazards models, and matched individuals without AKI or dialysis constituted the reference group. A Kaplan–Meier curve was generated from unadjusted hazards models. We further adjusted for age, sex, comorbidities, procedure codes and a propensity score in the multivariate models. Subsequent advanced chronic kidney disease (CKD) after discharge was evaluated for its effect on long-term severe sepsis using a time-dependent Cox analysis. The adjusted HR values for severe sepsis were further stratified according to comorbidities.
We used a propensity score approach to account for baseline differences between the AKI groups. Multiple logistic regressions were used to generate propensity scores, which were the predicted probabilities of AKI requiring dialysis. Covariates that were statistically significant in the unadjusted analysis were fit into the model, including age; ICU admission during the index hospitalization; admission frequency; use of mechanical ventilation; comorbidities before admission, such as diabetes mellitus, tumor with metastasis, CKD, chronic obstructive pulmonary disease and congestive heart failure; and comorbidities during index hospitalization, such as those of renal, metabolic, pulmonary or cardiovascular origin. We incorporated this score into a Cox proportional hazards model as a covariate.
In subgroup analysis, each acute dialysis patient with recovery from dialysis was matched with patients without AKI according to age, sex and history of diabetes. A two-sided P value less than 0.05 was considered statistically significant. All data were analyzed using R statistical software (version 2.14.1; Free Software Foundation, Inc, Boston, MA, USA).
Materials and Methods
Data Sources
The National Health Insurance (NHI) is a nationwide, compulsory, comprehensive health system in Taiwan. Patients were drawn from the NHI Research Database (NHIRD), which was released for research purposes by the National Health Research Institutes, Taipei, Taiwan. The NHIRD, one of the largest databases in the world, covers nearly all (99%) inpatient and outpatient claims for its population of more than 22 million people, and it has been used extensively in previous studies. The NHIRD provides encrypted patient identification numbers; age; sex; dates of hospital admission and discharge; medical institutions providing services; International Classification of Diseases, Ninth Edition, Clinical Modification (ICD-9-CM), codes of diagnoses and procedures; and outcome at hospital discharge (recovered, died or transferred out).
Patient Selection and Definition
The NHIRD consists of all the original claim data and registration files of one million individuals from 1999 to 2008. These one million individuals were randomly sampled from the 2000 Registry for Beneficiaries of the Taiwan NHI program and screening comorbidities by using the medical records for one year preceding admission. We linked to diagnostic codes through the hospitalization claims data to identify all episodes of AKI and the corresponding order codes used for the study subjects. We defined the first AKI hospitalization requiring dialysis as the index hospitalization for each individual. Participants who received dialysis during the index hospitalization were included. We excluded individuals who had a previous diagnosis of AKI, had received a kidney transplant or any form of dialysis preceding the index hospitalization AKI (n = 6,420), were hospitalized for more than 180 days with AKI (n = 36), died during the index hospitalization AKI (n = 718) or withdrew from the NHI within three months after discharge (n = 1,382). Each AKI patient was then matched with four hospitalized patients without AKI according to age, sex and history of diabetes (Figure 1). Informed consent was originally obtained by the NHRI, and, since patients were anonymous in the present study, informed consent was not required. Also, since the identification numbers of all individuals in the NHRID were encrypted to protect the privacy of the individuals, this study was exempt from a full ethical review by the institutional ethics review board of National Taiwan University Hospital (reference no. 201212021RINC).
Ascertainment of Study Outcome
Patients were tracked for outcomes beginning 90 days after discharge and ending at 31 December 2009. The main study outcome was severe sepsis. The method of identification of patients with severe sepsis was similar to that used by Angus et al., who selected all acute care hospitalizations with ICD-9-CM codes for both a bacterial or fungal infection process and a diagnosis of acute organ dysfunction. We used codes for acute organ dysfunction as modified by Shen et al. (listed in Table 1). Sepsis was defined according to the American College of Chest Physicians/Society of Critical Care Medicine (ACCP/SCCM) as systemic inflammatory syndrome in response to infection, which, when associated with acute organ dysfunction, is said to be severe. For patients with more than one hospital admission for severe sepsis during the study period, the first episode of sepsis was included.
Statistical Analyses
For individuals with and those without AKI and dialysis, continuous variables were compared using an unpaired t-test and are expressed as mean values with SD. Categorical variables were compared using a χ test and expressed as a percentage. The person-years were calculated from enrollment to outcome, death or the end of the study period. Incidence rates of severe sepsis were calculated for participants with and those without AKI and dialysis, respectively.
Each acute dialysis patient was matched with four non-AKI individuals according to age, sex and history of diabetes. Crude hazard ratios (HRs) with 95% confidence intervals (CIs) were derived from Cox proportional hazards models, and matched individuals without AKI or dialysis constituted the reference group. A Kaplan–Meier curve was generated from unadjusted hazards models. We further adjusted for age, sex, comorbidities, procedure codes and a propensity score in the multivariate models. Subsequent advanced chronic kidney disease (CKD) after discharge was evaluated for its effect on long-term severe sepsis using a time-dependent Cox analysis. The adjusted HR values for severe sepsis were further stratified according to comorbidities.
We used a propensity score approach to account for baseline differences between the AKI groups. Multiple logistic regressions were used to generate propensity scores, which were the predicted probabilities of AKI requiring dialysis. Covariates that were statistically significant in the unadjusted analysis were fit into the model, including age; ICU admission during the index hospitalization; admission frequency; use of mechanical ventilation; comorbidities before admission, such as diabetes mellitus, tumor with metastasis, CKD, chronic obstructive pulmonary disease and congestive heart failure; and comorbidities during index hospitalization, such as those of renal, metabolic, pulmonary or cardiovascular origin. We incorporated this score into a Cox proportional hazards model as a covariate.
In subgroup analysis, each acute dialysis patient with recovery from dialysis was matched with patients without AKI according to age, sex and history of diabetes. A two-sided P value less than 0.05 was considered statistically significant. All data were analyzed using R statistical software (version 2.14.1; Free Software Foundation, Inc, Boston, MA, USA).
Source...